Quantization-Aware Healing: a compressed, 4-bit model that outperforms its full-precision original

Automated excerpt

We introduce Quantization-Aware Healing (QAH), and applied to a GPT-OSS 120B model compressed to 60B parameters and quantized to MXFP4, it produces a model that beats its own full-precision (bfloat16) version on 7 of 9 benchmarks. This inverts the usual relationship between a 4-bit model and the 16-bit model it came from. The dominant healing recipe is quantization-aware training (QAT). Results We applied QAH to a GPT-OSS 120B model, compressed to 60B parameters and recovered in bfloat16, then re-quantized to MXFP4 under QAH. The QAH model wins on 7 of the 9 benchmarks.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.